This is a continuation of the post from Sunday, a smarter look at turnovers. Yesterday we looked at offenses, today we look at defenses. When referring to defenses, the points gained due to turnovers is represented as +TO instead of the -TO for offenses.
Turnovers are one of the more random outcomes in football (another being long field goals). For example, I can usually tell you how many yards per rush a team will get in a game in advance. I might be off by a yard, even two in the exceptional case, but I won't be off by five or six yards. Turnovers, on the other hand, are incredibly hard to predict. It doesn't surprise me at all when a team turns it over four times when the "should" have turned it over one or two. But I would be off by 100% to 300%. And those two or three turnovers will cost a team between 6 and 10 points in most cases.
Why is that? When things appear random it is because they are largely explained by events/conditions that we cannot measure or predict. For example, the flight of a bowling ball is explained by initial velocity and angle. The flight of a paper airplane by velocity, angle, and the coincidence of air patterns and position at various points in time, so the flight of the paper airplane is more random. Turnovers are the paper airplanes of football. So, while some teams force more turnovers and some teams turn the ball over more, turnovers are also about tips and ducks, hand and helmet placement, etc.
What this means is that, even over the course of an entire season let alone individual games, while I can be very confident that the team that has allowed few yards per carry has a good run defense, I am less confident that the team that has forced a lot of turnovers is good at creating turnovers. It is hard to know to what extent Fresno State and Kent State had good seasons because they lucked out with turnovers, or they were good at forcing turnovers and thus had good seasons. I am fairly confidence SMU was lucky to get to -103 since almost half of that came from two games. Boise State, Kansas State and, to a lesser extent, Oregon have made a living in their current incarnations off of +TO, so they don't surprise me.
At the other end, the Texas Tech defense always looked better on paper than in practice, and the +TO of -25 might have something to do with that. Since a high +TO is bad it is no surprise that Auburn scores pretty high. Somewhat surprising is that the nation's elite defenses do not score that well. LSU comes in at 23rd, Alabama at 33rd, Florida at 46th, and Notre Dame, despite the illustrious play of Stephon Tuitt (apparently Te'o is not actually on the field all by himself), is 53rd.
Curious if you have any estimates of how much of def. turnover rates is "luck" vs. "skill"? I know it is more skill (repeatable) in college than it is in NFL (almost all luck), based on a study I read somewhere, but I cannot seem to find it now.
ReplyDeleteIf you find that study let me know. I'd like to see their methodology. I can think of a dozen ways of testing this, but they all have serious flaws.
DeleteThe simplest approach is to calculate the turnover rate (turnovers per possession) for the first half and second half of the season and see if the two are correlated - if some teams are good at creating turnovers they should have a high rate in both halves, and vice versa.
For 2005-2012, you get a correlation of .2174. The result is significant so it's more than luck, but turnover rate in the first half only explains about 4-5% of the variation in the second half. My guess is that if you refined the methods you could probably get that up to about 15% skill, 85% luck.
Thanks. I don;t have the studies handy, but I know that def. INTs are much more unpredictable than off. INTs in the NFL. To the point where blindly fading the better defensive turnover producing team (above a certain threshold) has been a profitable strategy when betting ATS.
ReplyDelete